Skip to content
Back to search

paper-humanizer-skill

中英文学术文本润色与人性化skill,用于去除 AI 生成的痕迹,同时严格保持事实准确性。

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/crabin--paper-humanizer-skill/badge.svg)](/listings/crabin--paper-humanizer-skill)

HTML

<a href="/listings/crabin--paper-humanizer-skill"><img src="/api/listings/crabin--paper-humanizer-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 80 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: paper-humanizer
  3. description: Academic text polisher and humanizer for Chinese and English research writing. Removes AI-like patterns while preserving factual accuracy. Use when polishing academic text, reducing AI-generated patterns in papers/theses, improving naturalness of research writing while maintaining data integrity, or enhancing scholarly tone without fabricating facts. Supports both Chinese (中文) and English academic texts with configurable tone, field-specific terminology, and strict factuality controls.
  4. ---
  5. # Paper Humanizer
  6. Polish and humanize academic writing to reduce AI-like patterns while preserving all factual content, data, and meaning.
  7. ## Quick Start
  8. Apply the system prompt from `references/system_prompt.md` and use the workflow below.
  9. ## Output Format
  10. Always produce exactly 4 sections:
  11. 1. **原文 AI 特征分析** (Original Text AI Pattern Analysis)
  12. 2. **核心优化策略** (Core Optimization Strategies)
  13. 3. **优化亮点说明** (Highlight Notes - optional)
  14. 4. **优化后的文章** (Polished Text - full revised text)
  15. ## Core Constraints
  16. **Non-negotiable:**
  17. - Never fabricate facts, results, metrics, or references
  18. - Never change numeric values, experimental settings, or data
  19. - Preserve citation markers (e.g., [1], (Smith, 2023), \cite{...})
  20. - Maintain all terminology and technical accuracy
  21. **Remove AI-isms:**
  22. - Chinese: "值得注意的是", "不难发现", "基于以上分析", "综上所述", "首先/其次/最后", "本文将"
  23. - English: "It is worth noting that", "It can be seen that", "In summary", "Firstly, Secondly, Finally" (when overused)
  24. See `references/phrase_blacklist.md` for complete list.
  25. ## Parameters
  26. Configure behavior using these parameters:
  27. - **language**: `auto` (detect) | `zh` (Chinese) | `en` (English)
  28. - **tone**: `formal` (default) | `semi-formal` | `concise` | `persuasive`
  29. - **strict_factuality**: `true` (default) - preserve all data/facts
  30. - **keep_citations**: `true` (default) - preserve citation markers
  31. - **blacklist_level**: `high` (default) | `medium` | `low`
  32. ## Workflow
  33. 1. **Analyze**: Identify 2-3 major AI pattern issues in the input text
  34. 2. **Strategy**: Decide 3-5 targeted optimization strategies
  35. 3. **Execute**: Apply edits while strictly preserving factual content
  36. 4. **Output**: Deliver the 4-section format with full polished text
  37. ## Language Handling
  38. - Auto-detect language from input text
  39. - Maintain same language in output (unless user requests translation)
  40. - Keep technical terms (GAN, WGAN-GP, NSL-KDD) as-is
  41. - Preserve field-specific terminology
  42. ## Advanced Features
  43. For domain-specific terminology or special requirements, see:
  44. - `references/system_prompt.md` - Complete editing principles and constraints
  45. - `references/phrase_blacklist.md` - AI-pattern phrases to avoid
  46. - `references/user_template.md` - Parameterized user prompt template
  47. ## Script Usage (Optional)
  48. Use the CLI wrapper for batch processing:
  49. ```bash
  50. # Process a text file
  51. python3 scripts/paper_humanizer.py --text-file input.txt --language auto
  52. # Generate composed prompt for external use
  53. python3 scripts/paper_humanizer.py --text-file input.txt --out prompt.md
  54. ```
  55. The script combines system prompt and user template with specified parameters.
README.md · README · 273 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Paper Humanizer Skill
  2. 中英文学术文本润色与人性化Skill,用于去除 AI 生成的痕迹,同时严格保持事实准确性。
  3. ## ✨ 主要功能
  4. - 🎯 **去除 AI 痕迹**:识别并移除常见的 AI 写作模式
  5. - 📚 **学术语气优化**:保持学术严谨性,提升文本自然度
  6. - 🔒 **事实严格保护**:绝不编造数据、篡改数值或改变实验结论
  7. - 🌍 **双语支持**:完整支持中文和英文学术写作
  8. - ⚙️ **高度可配置**:多种参数配置,适应不同写作风格
  9. ## 🚀 快速开始
  10. ### 作为 Claude Code Skill 使用
  11. #### 安装步骤
  12. ```bash
  13. # 从 GitHub 克隆仓库
  14. git clone https://github.com/crabin/paper-humanizer-skill.git /tmp/paper-humanizer-skill
  15. # 创建 skills 目录(如果不存在)
  16. mkdir -p ~/.claude/skills/
  17. # 复制到 skills 目录
  18. cp -r /tmp/paper-humanizer-skill ~/.claude/skills/paper-humanizer
  19. # 验证安装
  20. ls ~/.claude/skills/paper-humanizer/
  21. ```
  22. #### 使用
  23. 在对话中直接使用,系统会自动应用相应的提示词。
  24. ### 命令行脚本使用
  25. ```bash
  26. # 处理文本文件
  27. python3 scripts/paper_humanizer.py --text-file input.txt --language auto
  28. # 生成组合提示词供外部使用
  29. python3 scripts/paper_humanizer.py --text-file input.txt --out prompt.md
  30. # 从标准输入读取
  31. cat input.txt | python3 scripts/paper_humanizer.py --language zh
  32. ```
  33. ## 📋 参数配置
  34. ### 基础参数
  35. | 参数 | 类型 | 默认值 | 说明 |
  36. |------|------|--------|------|
  37. | `--language` | auto/zh/en | auto | 文本语言(自动检测/中文/英文) |
  38. | `--tone` | formal/semi-formal/concise/persuasive | formal | 写作语气风格 |
  39. | `--field` | string | computer science | 研究领域(用于保持术语一致性) |
  40. | `--audience` | string | academic | 目标读者群体 |
  41. ### 约束控制
  42. | 参数 | 类型 | 默认值 | 说明 |
  43. |------|------|--------|------|
  44. | `--strict-factuality` | flag | True | 严格保持事实性(不篡改数据) |
  45. | `--no-strict-factuality` | flag | False | 允许适度调整表达 |
  46. | `--keep-citations` | flag | True | 保留引用标记 [1], (Smith, 2023) |
  47. | `--drop-citations` | flag | False | 移除引用标记 |
  48. | `--blacklist-level` | low/medium/high | high | AI 痍迹短语过滤级别 |
  49. ## 📝 输出格式
  50. 系统始终输出 4 个部分:
  51. ```
  52. 1. 原文 AI 特征分析:
  53. - 识别 2-3 个主要 AI 模式问题
  54. 2. 核心优化策略:
  55. - 列出 3-5 个针对性优化策略
  56. 3. 优化亮点说明:
  57. - 突出 1-2 个代表性修改(可选)
  58. 4. 优化后的文章:
  59. <完整优化后的文本>
  60. ```
  61. ## 🎯 核心原则
  62. ### 不可协商的约束
  63. - ❌ **绝不编造**新的事实、结果、指标或参考文献
  64. - ❌ **绝不改变**任何数值、超参数、阈值或对比结果
  65. - ✅ **保留所有**引用标记(如 [1]、(Smith, 2023)、\cite{...})
  66. - ✅ **保持术语**在目标领域的一致性
  67. ### 去除的 AI 痕迹
  68. **中文常见 AI 痕迹:**
  69. - "值得注意的是"、"不难发现"、"基于以上分析"
  70. - "综上所述"、"首先/其次/最后"、"本文将"
  71. - "在一定程度上"、"显著提升"(无量化支撑时)
  72. - "具有重要意义"、"此外"、"同时"(过度使用时)
  73. **英文常见 AI 痕迹:**
  74. - "It is worth noting that"、"It can be seen that"
  75. - "In summary"、"Firstly, Secondly, Finally"
  76. - "This paper aims to"、"To some extent"
  77. - "Significantly improves"(无证据时)
  78. 完整列表见 `references/phrase_blacklist.md`
  79. ## 📖 使用示例
  80. ### 中文示例
  81. **输入:**
  82. ```
  83. 值得注意的是,本文将提出一种基于WGAN-GP的改进方法,
  84. 并在NSL-KDD数据集上进行了实验。实验结果表明,该方法
  85. 在准确率上有显著提升。综上所述,本方法具有重要意义。
  86. ```
  87. **优化后:**
  88. ```
  89. 本研究提出了一种基于 WGAN-GP 的改进方法,并在 NSL-KDD
  90. 数据集上进行了验证。实验结果显示,该方法将准确率提升了
  91. X%,表明该方法在网络安全入侵检测领域具有应用价值。
  92. ```
  93. ### 英文示例
  94. **输入:**
  95. ```
  96. It is worth noting that this paper aims to propose an improved
  97. method based on WGAN-GP and conducts experiments on the NSL-KDD
  98. dataset. The experimental results show that the method significantly
  99. improves accuracy. In summary, the proposed method is of great significance.
  100. ```
  101. **优化后:**
  102. ```
  103. This study presents an improved WGAN-GP-based method, validated on
  104. the NSL-KDD dataset. Results demonstrate a X% accuracy improvement,
  105. indicating the method's practical value in network intrusion detection.
  106. ```
  107. ## 🧪 测试用例
  108. 以下两个测试用例可用于验证 skill 功能是否正常。
  109. ### 测试用例一:中文学术段落(含多种 AI 痕迹)
  110. **输入文本**(来自 `examples/zh_input_rich.txt`):
  111. ```
  112. 在当今人工智能技术迅猛发展的背景下,网络安全问题愈发凸显其重要性。值得注意的是,传统的入侵
  113. 检测系统在面对不断演变的网络攻击时,往往表现出明显的局限性。不难发现,现有方法在处理类别不
  114. 平衡数据集时存在诸多不足,严重制约了模型的泛化能力。
  115. 为了解决上述问题,本文旨在提出一种基于改进WGAN-GP的入侵检测方法。本文将深度学习与生成对抗
  116. 网络有机结合,充分发挥二者各自的优势,有效地、全面地解决了传统方法存在的不足之处。
  117. 综上所述,本文提出的方法不仅在技术层面实现了重要突破,更为入侵检测领域的发展提供了新的思路
  118. 和方向,具有重要的理论意义和实践价值。
  119. ```
  120. **预期 AI 特征分析**:
  121. - ❌ 模式词:「值得注意的是」「不难发现」「本文旨在」「综上所述」「有效地、全面地」「具有重要意义」
  122. - ❌ 空洞套话:「充分发挥」「有机结合」「具有重要的理论意义和实践价值」
  123. - ❌ 无量化支撑的「显著提升」「重要突破」
  124. **预期优化结果示例**:
  125. ```
  126. 传统入侵检测系统难以应对持续演变的网络攻击,在类别不平衡数据集上的泛化能力尤为薄弱。
  127. 本研究提出一种基于改进 WGAN-GP 的入侵检测方法,将生成对抗网络与深度学习相结合,针对
  128. 类别不平衡问题进行专项优化。实验结果表明,该方法在 NSL-KDD 数据集上的准确率、召回率
  129. 和 F1 值均有提升,具体改进幅度见表2。
  130. ```
  131. ---
  132. ### 测试用例二:英文学术段落(含多种 AI 痕迹)
  133. **输入文本**(来自 `examples/en_input_rich.txt`):
  134. ```
  135. In today's rapidly evolving digital landscape, cybersecurity threats are becoming increasingly
  136. sophisticated. It is worth noting that traditional intrusion detection systems often struggle
  137. to keep pace with the ever-changing nature of network attacks.
  138. To address this issue, this paper aims to propose an innovative framework based on an improved
  139. WGAN-GP architecture. Furthermore, the method holistically addresses the multifaceted challenges
  140. of intrusion detection in a comprehensive and robust manner.
  141. In conclusion, the proposed method represents a significant advancement in the field. Based on
  142. the above analysis, it can be concluded that this work has far-reaching implications for
  143. cybersecurity research and practice.
  144. ```
  145. **预期 AI 特征分析**:
  146. - ❌ 模式词:`It is worth noting that`、`this paper aims to`、`In conclusion`、`Based on the above analysis`
  147. - ❌ 空洞修饰:`rapidly evolving landscape`、`holistically addresses the multifaceted challenges`、`comprehensive and robust`
  148. - ❌ 无量化支撑的 `significant advancement`、`far-reaching implications`
  149. **预期优化结果示例**:
  150. ```
  151. Traditional intrusion detection systems struggle to adapt to evolving network attacks, particularly
  152. under class-imbalanced conditions. This work proposes an improved WGAN-GP framework that directly
  153. targets class imbalance in intrusion detection. On the NSL-KDD dataset, the method achieves
  154. improvements in accuracy, recall, and F1 score over existing baselines (see Table 2), suggesting
  155. practical applicability in real-world cybersecurity deployments.
  156. ```
  157. ---
  158. ## 📂 项目结构
  159. ```
  160. paper-humanizer/
  161. ├── README.md # 本文档
  162. ├── SKILL.md # Skill 元数据定义
  163. ├── references/ # 参考文档
  164. │ ├── system_prompt.md # 系统提示词(核心编辑原则)
  165. │ ├── user_template.md # 用户提示词模板
  166. │ └── phrase_blacklist.md # AI 痕迹短语黑名单
  167. ├── scripts/ # 脚本工具
  168. │ ├── paper_humanizer.py # Python CLI 工具
  169. │ └── paper_humanizer.sh # Bash 包装脚本
  170. └── examples/ # 使用示例
  171. ├── zh_input.txt # 中文简单示例输入
  172. ├── zh_input_rich.txt # 中文完整段落示例(含 6+ AI 痕迹)
  173. ├── en_input.txt # 英文简单示例输入
  174. ├── en_input_rich.txt # 英文完整段落示例(含 8+ AI 痕迹)
  175. └── run_demo.sh # 演示脚本
  176. ```
  177. ## ⚠️ 注意事项
  178. 1. **事实性保护**:如检测到原文存在矛盾或歧义,会在"核心优化策略"部分提出中性建议,而非直接修改文本
  179. 2. **技术术语**:保持技术术语原样(如 GAN、WGAN-GP、NSL-KDD),除非上下文要求标准翻译
  180. 3. **语言保持**:输出语言与输入保持一致,除非用户明确要求翻译
  181. 4. **引用标记**:默认保留所有引用标记,可通过 `--drop-citations` 参数移除
  182. ## 🔧 高级配置
  183. ### 领域特定术语
  184. 通过 `--field` 参数指定研究领域,系统会:
  185. - 保持该领域的术语一致性
  186. - 使用符合领域惯例的表达方式
  187. - 保留特定缩写和专业词汇
  188. ### 语气风格选择
  189. - `formal`:正式学术写作(论文、期刊投稿)
  190. - `semi-formal`:半正式(技术报告、博客)
  191. - `concise`:简洁风格(摘要、总结)
  192. - `persuasive`:说服性风格(申请材料、项目书)
  193. ## 📚 相关文档
  194. - `references/system_prompt.md` - 完整编辑原则和约束条件
  195. - `references/phrase_blacklist.md` - AI 痕迹短语完整列表
  196. - `references/user_template.md` - 参数化提示词模板
  197. - `SKILL.md` - Skill 元数据和快速参考
  198. ## 🤝 贡献
  199. 欢迎提交问题和改进建议!
  200. ## 📄 许可
  201. 本 skill 为学术辅助工具,请负责任地使用。优化后的文本仍需作者仔细校对。

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.